perf/x86: Fix time_shift in perf_event_mmap_page
authorAdrian Hunter <adrian.hunter@intel.com>
Fri, 16 Oct 2015 13:24:05 +0000 (16:24 +0300)
committerIngo Molnar <mingo@kernel.org>
Tue, 20 Oct 2015 08:30:52 +0000 (10:30 +0200)
Commit:

  b20112edeadf ("perf/x86: Improve accuracy of perf/sched clock")

allowed the time_shift value in perf_event_mmap_page to be as much
as 32.  Unfortunately the documented algorithms for using time_shift
have it shifting an integer, whereas to work correctly with the value
32, the type must be u64.

In the case of perf tools, Intel PT decodes correctly but the timestamps
that are output (for example by perf script) have lost 32-bits of
granularity so they look like they are not changing at all.

Fix by limiting the shift to 31 and adjusting the multiplier accordingly.

Also update the documentation of perf_event_mmap_page so that new code
based on it will be more future-proof.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Fixes: b20112edeadf ("perf/x86: Improve accuracy of perf/sched clock")
Link: http://lkml.kernel.org/r/1445001845-13688-2-git-send-email-adrian.hunter@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
arch/x86/kernel/tsc.c
include/uapi/linux/perf_event.h

index 69b84a26ea17a215bbd7169af0148b176b40f242..c7c4d9c51e99fe582b71ab1f2e3ca9ffaae2226c 100644 (file)
@@ -259,6 +259,17 @@ static void set_cyc2ns_scale(unsigned long cpu_khz, int cpu)
        clocks_calc_mult_shift(&data->cyc2ns_mul, &data->cyc2ns_shift, cpu_khz,
                               NSEC_PER_MSEC, 0);
 
+       /*
+        * cyc2ns_shift is exported via arch_perf_update_userpage() where it is
+        * not expected to be greater than 31 due to the original published
+        * conversion algorithm shifting a 32-bit value (now specifies a 64-bit
+        * value) - refer perf_event_mmap_page documentation in perf_event.h.
+        */
+       if (data->cyc2ns_shift == 32) {
+               data->cyc2ns_shift = 31;
+               data->cyc2ns_mul >>= 1;
+       }
+
        data->cyc2ns_offset = ns_now -
                mul_u64_u32_shr(tsc_now, data->cyc2ns_mul, data->cyc2ns_shift);
 
index 2881145cda86cda91621da082ebec83ae490f3c5..6c72e72e975ce1db61ff449a72aca101ef86d792 100644 (file)
@@ -476,7 +476,7 @@ struct perf_event_mmap_page {
         *   u64 delta;
         *
         *   quot = (cyc >> time_shift);
-        *   rem = cyc & ((1 << time_shift) - 1);
+        *   rem = cyc & (((u64)1 << time_shift) - 1);
         *   delta = time_offset + quot * time_mult +
         *              ((rem * time_mult) >> time_shift);
         *
@@ -507,7 +507,7 @@ struct perf_event_mmap_page {
         * And vice versa:
         *
         *   quot = cyc >> time_shift;
-        *   rem  = cyc & ((1 << time_shift) - 1);
+        *   rem  = cyc & (((u64)1 << time_shift) - 1);
         *   timestamp = time_zero + quot * time_mult +
         *               ((rem * time_mult) >> time_shift);
         */